Sparse coding and challenges for Bayesian models of the brain

Trappenberg, Thomas; Hollensen, Paul

doi:10.1017/s0140525x12002300

Cited by 5 publications

(2 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Noteworthily, different modes of top-down control can be characterized depending on how ''priors" are defined. A distinction can thus be drawn between ''empirical priors" that are learnt from past observations, and ''systemic priors" that correspond to built-in, and often domain-specific, biases (Clark, 2013; see also Hohwy, Roepstorff, & Friston, 2008;Trappenberg & Hollensen, 2013). In the present study, the correlation between the severity of ASD symptomatology in the domain of social interaction and lack of initial preference for the TFT strategy would suggest that social systemic priors are weak or altogether lacking in ASD.…”

Section: Discussionmentioning

confidence: 63%

Reduced sensitivity to social priors during action prediction in adults with autism spectrum disorders

et al. 2017

View full text Add to dashboard Cite

An extensive amount of evidence has documented a diminished ability to predict and understand other people's action in individuals with autism spectrum disorders (ASD). Recently, two theoretical accounts, the "Hypo-priors" and the "Aberrant precision" hypotheses, have suggested that attenuated Bayesian priors or an imbalance of the precision ascribed to sensory evidence relative to prior expectations may be responsible for the atypical perceptual experience and difficulties with action understanding in ASD. In the present study, we aimed to directly investigate whether difficulties in the appreciation of others' intentions can be accounted for by abnormal interaction between these two types of information: (i) the sensory evidence conveyed by movement kinematics, and (ii) the observer's expectations, acquired from past experience or derived from prior knowledge. To test this hypothesis, we contrasted the ability to infer Non-Social and Social intentions in adults with and without ASD, using a series of tasks in which both sensory evidence and prior expectations were manipulated. The results showed that attenuated effect of prior expectations in ASD individuals does not result from a generalized impairment in mentalizing, but one confined to social intentions. Attenuated priors in the social domain predicted the severity of clinical symptoms in the area of social interaction. Importantly, however, we found that reduced priors in the social domain could be compensated by ASD through observational learning, i.e. through deriving statistical regularities from observed behaviours. This capacity to balance reduced social expectations by learning inversely correlated with the severity of repetitive and stereotyped behaviours. Collectively, these findings suggest that adults with ASD exhibit a disturbance in the inferential mechanism that integrates sensory evidence into prior beliefs to produce accurate inferences about other people's intentions.

show abstract

Section: Discussionmentioning

confidence: 63%

Reduced sensitivity to social priors during action prediction in adults with autism spectrum disorders

et al. 2017

View full text Add to dashboard Cite

show abstract

“…However, notice that these empirical priors depend on the sensory data, that is, they actually are posteriors-which is not altered by the fact that they, just as in plain Bayesian inference (see section "Bayesian inference and the role of action in perception"), are fed back to become the new priors for the next inference cycle. In any case, they do not dissolve the just-mentioned criticism of Bayesian models, which is about first priors, that is, about priors that are independent of the sensory data (see also Trappenberg & Hollensen, 2013). First priors are relevant, simply because they form the starting point of the inference process.…”

Section: Empirical Priorsmentioning

confidence: 97%

Structural coding versus free-energy predictive coding

Helm

2015

Psychon Bull Rev

View full text Add to dashboard Cite

Focusing on visual perceptual organization, this article contrasts the free-energy (FE) version of predictive coding (a recent Bayesian approach) to structural coding (a long-standing representational approach). Both use freeenergy minimization as metaphor for processing in the brain, but their formal elaborations of this metaphor are fundamentally different. FE predictive coding formalizes it by minimization of prediction errors, whereas structural coding formalizes it by minimization of the descriptive complexity of predictions. Here, both sides are evaluated. A conclusion regarding competence is that FE predictive coding uses a powerful modeling technique, but that structural coding has more explanatory power. A conclusion regarding performance is that FE predictive coding-though more detailed in its account of neurophysiological data-provides a less compelling cognitive architecture than that of structural coding, which, for instance, supplies formal support for the computationally powerful role it attributes to neuronal synchronization.

show abstract

An Advanced Version of Cognitive Structural Realism

Beni

2019

Studies in Brain and Mind

View full text Add to dashboard Cite

Sparse coding and challenges for Bayesian models of the brain

Cited by 5 publications

References 7 publications

Reduced sensitivity to social priors during action prediction in adults with autism spectrum disorders

Reduced sensitivity to social priors during action prediction in adults with autism spectrum disorders

Structural coding versus free-energy predictive coding

An Advanced Version of Cognitive Structural Realism

Contact Info

Product

Resources

About